Uplift modeling for location-based online advertising

Abstract

As consumers spend substantial hours on their mobile phones, brick-and-mortar businesses need to reach out to customers online and transfer them to offline stores (O2O, online-to-offline). Widely-used location-based advertising is considered as effective in O2O advertising since the chance that a consumer visits stores crucially depends on the geographical attributes (e.g. distance). However, it is still unclear how effective the location-based targeting is due to the difficulty in measuring the net effect of advertising interventions on consumers. In this paper, we apply uplift modeling to a unique dataset extracted from randomized controlled trials (RCTs) conducted on location-based advertising campaigns to find important geo-features and to predict user segments with high net effect of advertisement.